Researchers from Microsoft have achieved equal accuracy in the field of language acquisition with artificial intelligence or AI.
Recently, a team of Microsoft researchers said that they used the machine transcriptor Ai to achieve equally accurate skills in Chinese and English languages.
Researchers from the Asia and US labs of the organization said that their translation system has achieved equal human skills in the examination of 'Newslett 2017'. In this development, the industry related professionals and academic partners participated in the development of the translation system. WMT 17 - Results of the translation system are published at the research conference, according to a notification, the company said.
In order to verify the accuracy of this result, interpreter evaluators were appointed from outside, who compared the outcome of Microsoft's results to the results of comparative evaluation.
Arul Menezes, Partner Research Manager, Microsoft Machine Translation Team / Microsoft
Arul Menezes, Partner Research Manager, Microsoft Machine Translation Team / Microsoft
Chief of the National Language Processing Group, working at the project and Assistant Managing Director of Microsoft Research Asia, Ming Zhou said, "We are excited to achieve the milestone of achieving the equivalent of people in the dataset. However, there are still many adversities in this regard, such as verification of results in real-time news stories. "
Researchers have been working on translation for many years. However, they recently achieved significant success using the training AI system titled Deep Neural Networks, said Microsoft. This system is known as 'Statistic Machine Translation'.
In addition to adding numbers to achieve the milestone of achieving human equivalent in the dataset, three Microsoft researchers worked together in Beijing and Washington's Redmond to make the translation system more accurate by using training methods. In many cases, the method will follow the pronunciation of the person until it is absolutely correct.
Microsoft has used another method in machine translation, that is, dual-learning. It works like a kind of data verification. Each time the system gives a sentence for translation from Chinese language to English, the researcher translates sentences again from English to Chinese. People usually use this method to verify the accuracy of the translation case, the organization said. In addition to this, it will allow the system to get itself wrong. Microsoft's research team has also developed dual learning methods. This innovation can be used with artificial intelligence for other purposes.
Ti-Yan Liu, Principal Research Manager with Microsoft Research Asia, Beijing / Microsoft
Ti-Yan Liu, Principal Research Manager with Microsoft Research Asia, Beijing / Microsoft
Another method is, Deliverance Networks. It is like editing and modifying human writings by reading it manually. The researchers have taught the machine how to repeat the same sentence in the field of translation until translation is defective.
Joint training techniques have also been used in English translation from English to Chinese and Chinese to English translation in machine translation. This method is used to translate new English sentences from English to Chinese language translations and to pair with the new sentence in Chinese. This is the same process from English to Chinese translation. This will increase efficiency in two types of translations.
In this case, a system called 'Agreement Regulatory' has also been used. Through this method the system can translate from left to right or right to the left. If this technique is used in the same translation, it makes the translation results credible. This technique is very effective in maintaining proper rhythm of translation.
This post has received a 0.31 % upvote from @booster thanks to: @forhadgt.